Generative downscaling of PDE solvers with physics-guided diffusion models

Yulong Lu,Wuzhe Xu

2024-04-08

Abstract:Solving partial differential equations (PDEs) on fine spatio-temporal scales for high-fidelity solutions is critical for numerous scientific breakthroughs. Yet, this process can be prohibitively expensive, owing to the inherent complexities of the problems, including nonlinearity and multiscale phenomena. To speed up large-scale computations, a process known as downscaling is employed, which generates high-fidelity approximate solutions from their low-fidelity counterparts. In this paper, we propose a novel Physics-Guided Diffusion Model (PGDM) for downscaling. Our model, initially trained on a dataset comprising low-and-high-fidelity paired solutions across coarse and fine scales, generates new high-fidelity approximations from any new low-fidelity inputs. These outputs are subsequently refined through fine-tuning, aimed at minimizing the physical discrepancies as defined by the discretized PDEs at the finer scale. We evaluate and benchmark our model's performance against other downscaling baselines in three categories of nonlinear PDEs. Our numerical experiments demonstrate that our model not only outperforms the baselines but also achieves a computational acceleration exceeding tenfold, while maintaining the same level of accuracy as the conventional fine-scale solvers.

Numerical Analysis

What problem does this paper attempt to address?

This paper proposes an efficient method for solving partial differential equations (PDEs), especially for fast computation of high-fidelity solutions. In many scientific breakthroughs, accurate solving of PDEs on fine spatiotemporal scales is crucial, but it can be extremely expensive due to the complexity of the problem, such as nonlinear and multiscale phenomena. To address this issue, the paper introduces a new framework called Physics-Guided Diffusion Model (PGDM) for downsampling PDE solvers.

PGDM first trains on a dataset of solution pairs at low and high resolutions, and then generates high-resolution approximations from new low-resolution inputs, minimizing the physical error defined by the discretized PDE at fine scales through fine-tuning. This approach is compared with other downsampling baselines on three different types of nonlinear PDEs, demonstrating superior performance over the baselines while achieving more than ten times faster computation speed and maintaining comparable accuracy to traditional fine-scale solvers.

The main contributions of the paper are:
1. Reformulating the downsampling problem as a conditional sampling task to enhance the accuracy of generating high-resolution outputs.
2. Using a diffusion model for initial conditional sampling, followed by refinement through minimizing physics-informed loss to ensure adherence to the physical laws defined by the PDE.
3. Experimental results show that the proposed PGDM method performs excellently in various nonlinear static and time-varying PDEs, improving computational efficiency while maintaining accuracy.

In summary, the paper aims to accelerate the computation of PDEs, particularly for climate modeling and super-resolution imaging, by combining data-driven sample generation and physics-based accuracy enhancement, providing an effective approach for addressing the transformation between high and low resolutions.